Content
96%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A strong, evidence-grounded skill body: dense domain knowledge Claude does not have, fully executable commands, a consent-gated workflow with explicit error statuses, and honest limits/stop-rule sections. The one real defect is that the referenced `references/papers/` directory (including `INDEX.md`, and the per-paper files that `library.md` also points to) is absent from the bundle, so the paper-verification workflow starts from a broken path — partially mitigated by the in-body fallback instruction.
Suggestions
Ship the `references/papers/` directory with `INDEX.md` (and the open-access full-text files that `library.md` already cites by filename, e.g. `Racinais_2023_IOC-consensus-sport-events-in-the-heat.md`), so the 'Papers' verification workflow has a working starting point instead of relying entirely on the `paper_fetch.py` fallback.
If the paper files are intentionally not bundled, change the Papers section to say so explicitly (e.g. 'papers live in the `rampwise` skill's data directory; fetch via `paper_fetch.py`') rather than pointing at a `references/papers/` path that does not exist.
State where the athlete's `profile.md` (used for the `weather_lookup_approved` consent check) is found — `$RAMPWISE_DATA` is defined for personal data, but the consent check path `rampwise/profile.md` is not tied to that variable, leaving ambiguity about which location to read.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and every token earns its place: it carries only domain-specific knowledge Claude cannot already know (verified effect sizes from El Helou 2012, Ely 2007, Tatterson 2000, Daanen 2018 decay rates, the "no universal WBGT cut-off" finding) plus executable commands, with zero padding and no explanation of general concepts. Not 4: there is no over-explanation to trim — even the framing sentence about heat as the biggest HR confounder states the skill's core comparison rule rather than teaching known background. | 5 / 5 |
Actionability | Fully copy-paste-ready commands cover the common cases: `weather_check.py --activity-id ACTIVITY_ID --approved [--json]`, `acclimation_decay.py --days-since-heat 14`, `weather.py --lat ... --date ... --start ... --hours 2`, and `swim_water.py --temp-c <water> [--km <distance>]`, each with its real flags explained. Not 4: the guidance goes beyond commands to a concrete consent script to speak aloud, a specified handoff output block with field-level detail, and a stop rule with exact symptoms and actions. | 5 / 5 |
Workflow Clarity | The sequence is clear with an explicit validation gate for the one risky operation (the external weather lookup): check `rampwise/profile.md` for `weather_lookup_approved`, ask once with a given script, record yes/no, and "Without it the script refuses and sends nothing" — a hard checkpoint with feedback. Error recovery is enumerated in the handoff block (STATUS: ok | no-consent | no-position | lookup-failed) and the Papers section adds a verify-then-fallback loop for missing or abstract-only sources. Not 4: validation checkpoints and failure paths are explicit at every stage, not just present. | 5 / 5 |
Progressive Disclosure | Good structure overall: SKILL.md is an overview with clear sections, the evidence is one level deep and clearly signaled ("evidence in `references/evidence-map.md`" in a section header), and all five referenced scripts exist in `scripts/`. However, the Papers section points to `references/papers/` with an `INDEX.md` "lists the read depth", and that directory does not exist in the bundle — a broken navigation path. Not 5: easy navigation is undermined by the dangling `references/papers/INDEX.md` reference; not 3: references are clearly signaled, content is appropriately split rather than inlined, and the body anticipates missing paper files with an explicit fallback (run `paper_fetch.py` from the `rampwise` skill), keeping this a minor gap. | 4 / 5 |
Total | 19 / 20 Passed |